src/execute.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU General Public License as published by
  10   the Free Software Foundation; either version 2 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   General Public License for more details.
  17
  18   You should have received a copy of the GNU General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <assert.h>
  23 #include <dirent.h>
  24 #include <errno.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include <string.h>
  28 #include <signal.h>
  29 #include <sys/socket.h>
  30 #include <sys/un.h>
  31 #include <sys/prctl.h>
  32 #include <linux/sched.h>
  33 #include <sys/types.h>
  34 #include <sys/stat.h>
  35 #include <grp.h>
  36 #include <pwd.h>
  37 #include <sys/mount.h>
  38 #include <linux/fs.h>
  39 #include <linux/oom.h>
  40
  41 #ifdef HAVE_PAM
  42 #include <security/pam_appl.h>
  43 #endif
  44
  45 #include "execute.h"
  46 #include "strv.h"
  47 #include "macro.h"
  48 #include "util.h"
  49 #include "log.h"
  50 #include "ioprio.h"
  51 #include "securebits.h"
  52 #include "cgroup.h"
  53 #include "namespace.h"
  54 #include "tcpwrap.h"
  55 #include "exit-status.h"
  56 #include "missing.h"
  57 #include "utmp-wtmp.h"
  58 #include "def.h"
  59 #include "loopback-setup.h"
  60
  61 /* This assumes there is a 'tty' group */
  62 #define TTY_MODE 0620
  63
  64 static int shift_fds(int fds[], unsigned n_fds) {
  65         int start, restart_from;
  66
  67         if (n_fds <= 0)
  68                 return 0;
  69
  70         /* Modifies the fds array! (sorts it) */
  71
  72         assert(fds);
  73
  74         start = 0;
  75         for (;;) {
  76                 int i;
  77
  78                 restart_from = -1;
  79
  80                 for (i = start; i < (int) n_fds; i++) {
  81                         int nfd;
  82
  83                         /* Already at right index? */
  84                         if (fds[i] == i+3)
  85                                 continue;
  86
  87                         if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
  88                                 return -errno;
  89
  90                         close_nointr_nofail(fds[i]);
  91                         fds[i] = nfd;
  92
  93                         /* Hmm, the fd we wanted isn't free? Then
  94                          * let's remember that and try again from here*/
  95                         if (nfd != i+3 && restart_from < 0)
  96                                 restart_from = i;
  97                 }
  98
  99                 if (restart_from < 0)
 100                         break;
 101
 102                 start = restart_from;
 103         }
 104
 105         return 0;
 106 }
 107
 108 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
 109         unsigned i;
 110         int r;
 111
 112         if (n_fds <= 0)
 113                 return 0;
 114
 115         assert(fds);
 116
 117         /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
 118
 119         for (i = 0; i < n_fds; i++) {
 120
 121                 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
 122                         return r;
 123
 124                 /* We unconditionally drop FD_CLOEXEC from the fds,
 125                  * since after all we want to pass these fds to our
 126                  * children */
 127
 128                 if ((r = fd_cloexec(fds[i], false)) < 0)
 129                         return r;
 130         }
 131
 132         return 0;
 133 }
 134
 135 static const char *tty_path(const ExecContext *context) {
 136         assert(context);
 137
 138         if (context->tty_path)
 139                 return context->tty_path;
 140
 141         return "/dev/console";
 142 }
 143
 144 void exec_context_tty_reset(const ExecContext *context) {
 145         assert(context);
 146
 147         if (context->tty_vhangup)
 148                 terminal_vhangup(tty_path(context));
 149
 150         if (context->tty_reset)
 151                 reset_terminal(tty_path(context));
 152
 153         if (context->tty_vt_disallocate && context->tty_path)
 154                 vt_disallocate(context->tty_path);
 155 }
 156
 157 static int open_null_as(int flags, int nfd) {
 158         int fd, r;
 159
 160         assert(nfd >= 0);
 161
 162         if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
 163                 return -errno;
 164
 165         if (fd != nfd) {
 166                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 167                 close_nointr_nofail(fd);
 168         } else
 169                 r = nfd;
 170
 171         return r;
 172 }
 173
 174 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
 175         int fd, r;
 176         union sockaddr_union sa;
 177
 178         assert(context);
 179         assert(output < _EXEC_OUTPUT_MAX);
 180         assert(ident);
 181         assert(nfd >= 0);
 182
 183         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 184         if (fd < 0)
 185                 return -errno;
 186
 187         zero(sa);
 188         sa.un.sun_family = AF_UNIX;
 189         strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path));
 190
 191         r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
 192         if (r < 0) {
 193                 close_nointr_nofail(fd);
 194                 return -errno;
 195         }
 196
 197         if (shutdown(fd, SHUT_RD) < 0) {
 198                 close_nointr_nofail(fd);
 199                 return -errno;
 200         }
 201
 202         dprintf(fd,
 203                 "%s\n"
 204                 "%i\n"
 205                 "%i\n"
 206                 "%i\n"
 207                 "%i\n"
 208                 "%i\n",
 209                 context->syslog_identifier ? context->syslog_identifier : ident,
 210                 context->syslog_priority,
 211                 !!context->syslog_level_prefix,
 212                 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
 213                 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
 214                 output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || output == EXEC_OUTPUT_KMSG_AND_CONSOLE);
 215
 216         if (fd != nfd) {
 217                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 218                 close_nointr_nofail(fd);
 219         } else
 220                 r = nfd;
 221
 222         return r;
 223 }
 224 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
 225         int fd, r;
 226
 227         assert(path);
 228         assert(nfd >= 0);
 229
 230         if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
 231                 return fd;
 232
 233         if (fd != nfd) {
 234                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
 235                 close_nointr_nofail(fd);
 236         } else
 237                 r = nfd;
 238
 239         return r;
 240 }
 241
 242 static bool is_terminal_input(ExecInput i) {
 243         return
 244                 i == EXEC_INPUT_TTY ||
 245                 i == EXEC_INPUT_TTY_FORCE ||
 246                 i == EXEC_INPUT_TTY_FAIL;
 247 }
 248
 249 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
 250
 251         if (is_terminal_input(std_input) && !apply_tty_stdin)
 252                 return EXEC_INPUT_NULL;
 253
 254         if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
 255                 return EXEC_INPUT_NULL;
 256
 257         return std_input;
 258 }
 259
 260 static int fixup_output(ExecOutput std_output, int socket_fd) {
 261
 262         if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
 263                 return EXEC_OUTPUT_INHERIT;
 264
 265         return std_output;
 266 }
 267
 268 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
 269         ExecInput i;
 270
 271         assert(context);
 272
 273         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 274
 275         switch (i) {
 276
 277         case EXEC_INPUT_NULL:
 278                 return open_null_as(O_RDONLY, STDIN_FILENO);
 279
 280         case EXEC_INPUT_TTY:
 281         case EXEC_INPUT_TTY_FORCE:
 282         case EXEC_INPUT_TTY_FAIL: {
 283                 int fd, r;
 284
 285                 if ((fd = acquire_terminal(
 286                                      tty_path(context),
 287                                      i == EXEC_INPUT_TTY_FAIL,
 288                                      i == EXEC_INPUT_TTY_FORCE,
 289                                      false)) < 0)
 290                         return fd;
 291
 292                 if (fd != STDIN_FILENO) {
 293                         r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 294                         close_nointr_nofail(fd);
 295                 } else
 296                         r = STDIN_FILENO;
 297
 298                 return r;
 299         }
 300
 301         case EXEC_INPUT_SOCKET:
 302                 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
 303
 304         default:
 305                 assert_not_reached("Unknown input type");
 306         }
 307 }
 308
 309 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 310         ExecOutput o;
 311         ExecInput i;
 312
 313         assert(context);
 314         assert(ident);
 315
 316         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 317         o = fixup_output(context->std_output, socket_fd);
 318
 319         /* This expects the input is already set up */
 320
 321         switch (o) {
 322
 323         case EXEC_OUTPUT_INHERIT:
 324
 325                 /* If input got downgraded, inherit the original value */
 326                 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
 327                         return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 328
 329                 /* If the input is connected to anything that's not a /dev/null, inherit that... */
 330                 if (i != EXEC_INPUT_NULL)
 331                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 332
 333                 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
 334                 if (getppid() != 1)
 335                         return STDOUT_FILENO;
 336
 337                 /* We need to open /dev/null here anew, to get the
 338                  * right access mode. So we fall through */
 339
 340         case EXEC_OUTPUT_NULL:
 341                 return open_null_as(O_WRONLY, STDOUT_FILENO);
 342
 343         case EXEC_OUTPUT_TTY:
 344                 if (is_terminal_input(i))
 345                         return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 346
 347                 /* We don't reset the terminal if this is just about output */
 348                 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
 349
 350         case EXEC_OUTPUT_SYSLOG:
 351         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 352         case EXEC_OUTPUT_KMSG:
 353         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 354                 return connect_logger_as(context, o, ident, STDOUT_FILENO);
 355
 356         case EXEC_OUTPUT_SOCKET:
 357                 assert(socket_fd >= 0);
 358                 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
 359
 360         default:
 361                 assert_not_reached("Unknown output type");
 362         }
 363 }
 364
 365 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
 366         ExecOutput o, e;
 367         ExecInput i;
 368
 369         assert(context);
 370         assert(ident);
 371
 372         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
 373         o = fixup_output(context->std_output, socket_fd);
 374         e = fixup_output(context->std_error, socket_fd);
 375
 376         /* This expects the input and output are already set up */
 377
 378         /* Don't change the stderr file descriptor if we inherit all
 379          * the way and are not on a tty */
 380         if (e == EXEC_OUTPUT_INHERIT &&
 381             o == EXEC_OUTPUT_INHERIT &&
 382             i == EXEC_INPUT_NULL &&
 383             !is_terminal_input(context->std_input) &&
 384             getppid () != 1)
 385                 return STDERR_FILENO;
 386
 387         /* Duplicate from stdout if possible */
 388         if (e == o || e == EXEC_OUTPUT_INHERIT)
 389                 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 390
 391         switch (e) {
 392
 393         case EXEC_OUTPUT_NULL:
 394                 return open_null_as(O_WRONLY, STDERR_FILENO);
 395
 396         case EXEC_OUTPUT_TTY:
 397                 if (is_terminal_input(i))
 398                         return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 399
 400                 /* We don't reset the terminal if this is just about output */
 401                 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
 402
 403         case EXEC_OUTPUT_SYSLOG:
 404         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
 405         case EXEC_OUTPUT_KMSG:
 406         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
 407                 return connect_logger_as(context, e, ident, STDERR_FILENO);
 408
 409         case EXEC_OUTPUT_SOCKET:
 410                 assert(socket_fd >= 0);
 411                 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
 412
 413         default:
 414                 assert_not_reached("Unknown error type");
 415         }
 416 }
 417
 418 static int chown_terminal(int fd, uid_t uid) {
 419         struct stat st;
 420
 421         assert(fd >= 0);
 422
 423         /* This might fail. What matters are the results. */
 424         (void) fchown(fd, uid, -1);
 425         (void) fchmod(fd, TTY_MODE);
 426
 427         if (fstat(fd, &st) < 0)
 428                 return -errno;
 429
 430         if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
 431                 return -EPERM;
 432
 433         return 0;
 434 }
 435
 436 static int setup_confirm_stdio(const ExecContext *context,
 437                                int *_saved_stdin,
 438                                int *_saved_stdout) {
 439         int fd = -1, saved_stdin, saved_stdout = -1, r;
 440
 441         assert(context);
 442         assert(_saved_stdin);
 443         assert(_saved_stdout);
 444
 445         /* This returns positive EXIT_xxx return values instead of
 446          * negative errno style values! */
 447
 448         if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
 449                 return EXIT_STDIN;
 450
 451         if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
 452                 r = EXIT_STDOUT;
 453                 goto fail;
 454         }
 455
 456         if ((fd = acquire_terminal(
 457                              tty_path(context),
 458                              context->std_input == EXEC_INPUT_TTY_FAIL,
 459                              context->std_input == EXEC_INPUT_TTY_FORCE,
 460                              false)) < 0) {
 461                 r = EXIT_STDIN;
 462                 goto fail;
 463         }
 464
 465         if (chown_terminal(fd, getuid()) < 0) {
 466                 r = EXIT_STDIN;
 467                 goto fail;
 468         }
 469
 470         if (dup2(fd, STDIN_FILENO) < 0) {
 471                 r = EXIT_STDIN;
 472                 goto fail;
 473         }
 474
 475         if (dup2(fd, STDOUT_FILENO) < 0) {
 476                 r = EXIT_STDOUT;
 477                 goto fail;
 478         }
 479
 480         if (fd >= 2)
 481                 close_nointr_nofail(fd);
 482
 483         *_saved_stdin = saved_stdin;
 484         *_saved_stdout = saved_stdout;
 485
 486         return 0;
 487
 488 fail:
 489         if (saved_stdout >= 0)
 490                 close_nointr_nofail(saved_stdout);
 491
 492         if (saved_stdin >= 0)
 493                 close_nointr_nofail(saved_stdin);
 494
 495         if (fd >= 0)
 496                 close_nointr_nofail(fd);
 497
 498         return r;
 499 }
 500
 501 static int restore_confirm_stdio(const ExecContext *context,
 502                                  int *saved_stdin,
 503                                  int *saved_stdout,
 504                                  bool *keep_stdin,
 505                                  bool *keep_stdout) {
 506
 507         assert(context);
 508         assert(saved_stdin);
 509         assert(*saved_stdin >= 0);
 510         assert(saved_stdout);
 511         assert(*saved_stdout >= 0);
 512
 513         /* This returns positive EXIT_xxx return values instead of
 514          * negative errno style values! */
 515
 516         if (is_terminal_input(context->std_input)) {
 517
 518                 /* The service wants terminal input. */
 519
 520                 *keep_stdin = true;
 521                 *keep_stdout =
 522                         context->std_output == EXEC_OUTPUT_INHERIT ||
 523                         context->std_output == EXEC_OUTPUT_TTY;
 524
 525         } else {
 526                 /* If the service doesn't want a controlling terminal,
 527                  * then we need to get rid entirely of what we have
 528                  * already. */
 529
 530                 if (release_terminal() < 0)
 531                         return EXIT_STDIN;
 532
 533                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
 534                         return EXIT_STDIN;
 535
 536                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
 537                         return EXIT_STDOUT;
 538
 539                 *keep_stdout = *keep_stdin = false;
 540         }
 541
 542         return 0;
 543 }
 544
 545 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
 546         bool keep_groups = false;
 547         int r;
 548
 549         assert(context);
 550
 551         /* Lookup and set GID and supplementary group list. Here too
 552          * we avoid NSS lookups for gid=0. */
 553
 554         if (context->group || username) {
 555
 556                 if (context->group) {
 557                         const char *g = context->group;
 558
 559                         if ((r = get_group_creds(&g, &gid)) < 0)
 560                                 return r;
 561                 }
 562
 563                 /* First step, initialize groups from /etc/groups */
 564                 if (username && gid != 0) {
 565                         if (initgroups(username, gid) < 0)
 566                                 return -errno;
 567
 568                         keep_groups = true;
 569                 }
 570
 571                 /* Second step, set our gids */
 572                 if (setresgid(gid, gid, gid) < 0)
 573                         return -errno;
 574         }
 575
 576         if (context->supplementary_groups) {
 577                 int ngroups_max, k;
 578                 gid_t *gids;
 579                 char **i;
 580
 581                 /* Final step, initialize any manually set supplementary groups */
 582                 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
 583
 584                 if (!(gids = new(gid_t, ngroups_max)))
 585                         return -ENOMEM;
 586
 587                 if (keep_groups) {
 588                         if ((k = getgroups(ngroups_max, gids)) < 0) {
 589                                 free(gids);
 590                                 return -errno;
 591                         }
 592                 } else
 593                         k = 0;
 594
 595                 STRV_FOREACH(i, context->supplementary_groups) {
 596                         const char *g;
 597
 598                         if (k >= ngroups_max) {
 599                                 free(gids);
 600                                 return -E2BIG;
 601                         }
 602
 603                         g = *i;
 604                         r = get_group_creds(&g, gids+k);
 605                         if (r < 0) {
 606                                 free(gids);
 607                                 return r;
 608                         }
 609
 610                         k++;
 611                 }
 612
 613                 if (setgroups(k, gids) < 0) {
 614                         free(gids);
 615                         return -errno;
 616                 }
 617
 618                 free(gids);
 619         }
 620
 621         return 0;
 622 }
 623
 624 static int enforce_user(const ExecContext *context, uid_t uid) {
 625         int r;
 626         assert(context);
 627
 628         /* Sets (but doesn't lookup) the uid and make sure we keep the
 629          * capabilities while doing so. */
 630
 631         if (context->capabilities) {
 632                 cap_t d;
 633                 static const cap_value_t bits[] = {
 634                         CAP_SETUID,   /* Necessary so that we can run setresuid() below */
 635                         CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
 636                 };
 637
 638                 /* First step: If we need to keep capabilities but
 639                  * drop privileges we need to make sure we keep our
 640                  * caps, whiel we drop privileges. */
 641                 if (uid != 0) {
 642                         int sb = context->secure_bits|SECURE_KEEP_CAPS;
 643
 644                         if (prctl(PR_GET_SECUREBITS) != sb)
 645                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
 646                                         return -errno;
 647                 }
 648
 649                 /* Second step: set the capabilities. This will reduce
 650                  * the capabilities to the minimum we need. */
 651
 652                 if (!(d = cap_dup(context->capabilities)))
 653                         return -errno;
 654
 655                 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
 656                     cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
 657                         r = -errno;
 658                         cap_free(d);
 659                         return r;
 660                 }
 661
 662                 if (cap_set_proc(d) < 0) {
 663                         r = -errno;
 664                         cap_free(d);
 665                         return r;
 666                 }
 667
 668                 cap_free(d);
 669         }
 670
 671         /* Third step: actually set the uids */
 672         if (setresuid(uid, uid, uid) < 0)
 673                 return -errno;
 674
 675         /* At this point we should have all necessary capabilities but
 676            are otherwise a normal user. However, the caps might got
 677            corrupted due to the setresuid() so we need clean them up
 678            later. This is done outside of this call. */
 679
 680         return 0;
 681 }
 682
 683 #ifdef HAVE_PAM
 684
 685 static int null_conv(
 686                 int num_msg,
 687                 const struct pam_message **msg,
 688                 struct pam_response **resp,
 689                 void *appdata_ptr) {
 690
 691         /* We don't support conversations */
 692
 693         return PAM_CONV_ERR;
 694 }
 695
 696 static int setup_pam(
 697                 const char *name,
 698                 const char *user,
 699                 const char *tty,
 700                 char ***pam_env,
 701                 int fds[], unsigned n_fds) {
 702
 703         static const struct pam_conv conv = {
 704                 .conv = null_conv,
 705                 .appdata_ptr = NULL
 706         };
 707
 708         pam_handle_t *handle = NULL;
 709         sigset_t ss, old_ss;
 710         int pam_code = PAM_SUCCESS;
 711         int err;
 712         char **e = NULL;
 713         bool close_session = false;
 714         pid_t pam_pid = 0, parent_pid;
 715
 716         assert(name);
 717         assert(user);
 718         assert(pam_env);
 719
 720         /* We set up PAM in the parent process, then fork. The child
 721          * will then stay around until killed via PR_GET_PDEATHSIG or
 722          * systemd via the cgroup logic. It will then remove the PAM
 723          * session again. The parent process will exec() the actual
 724          * daemon. We do things this way to ensure that the main PID
 725          * of the daemon is the one we initially fork()ed. */
 726
 727         if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
 728                 handle = NULL;
 729                 goto fail;
 730         }
 731
 732         if (tty)
 733                 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
 734                         goto fail;
 735
 736         if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
 737                 goto fail;
 738
 739         if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
 740                 goto fail;
 741
 742         close_session = true;
 743
 744         if ((!(e = pam_getenvlist(handle)))) {
 745                 pam_code = PAM_BUF_ERR;
 746                 goto fail;
 747         }
 748
 749         /* Block SIGTERM, so that we know that it won't get lost in
 750          * the child */
 751         if (sigemptyset(&ss) < 0 ||
 752             sigaddset(&ss, SIGTERM) < 0 ||
 753             sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
 754                 goto fail;
 755
 756         parent_pid = getpid();
 757
 758         if ((pam_pid = fork()) < 0)
 759                 goto fail;
 760
 761         if (pam_pid == 0) {
 762                 int sig;
 763                 int r = EXIT_PAM;
 764
 765                 /* The child's job is to reset the PAM session on
 766                  * termination */
 767
 768                 /* This string must fit in 10 chars (i.e. the length
 769                  * of "/sbin/init") */
 770                 rename_process("sd(PAM)");
 771
 772                 /* Make sure we don't keep open the passed fds in this
 773                 child. We assume that otherwise only those fds are
 774                 open here that have been opened by PAM. */
 775                 close_many(fds, n_fds);
 776
 777                 /* Wait until our parent died. This will most likely
 778                  * not work since the kernel does not allow
 779                  * unprivileged parents kill their privileged children
 780                  * this way. We rely on the control groups kill logic
 781                  * to do the rest for us. */
 782                 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
 783                         goto child_finish;
 784
 785                 /* Check if our parent process might already have
 786                  * died? */
 787                 if (getppid() == parent_pid) {
 788                         for (;;) {
 789                                 if (sigwait(&ss, &sig) < 0) {
 790                                         if (errno == EINTR)
 791                                                 continue;
 792
 793                                         goto child_finish;
 794                                 }
 795
 796                                 assert(sig == SIGTERM);
 797                                 break;
 798                         }
 799                 }
 800
 801                 /* If our parent died we'll end the session */
 802                 if (getppid() != parent_pid)
 803                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
 804                                 goto child_finish;
 805
 806                 r = 0;
 807
 808         child_finish:
 809                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 810                 _exit(r);
 811         }
 812
 813         /* If the child was forked off successfully it will do all the
 814          * cleanups, so forget about the handle here. */
 815         handle = NULL;
 816
 817         /* Unblock SIGTERM again in the parent */
 818         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
 819                 goto fail;
 820
 821         /* We close the log explicitly here, since the PAM modules
 822          * might have opened it, but we don't want this fd around. */
 823         closelog();
 824
 825         *pam_env = e;
 826         e = NULL;
 827
 828         return 0;
 829
 830 fail:
 831         if (pam_code != PAM_SUCCESS)
 832                 err = -EPERM;  /* PAM errors do not map to errno */
 833         else
 834                 err = -errno;
 835
 836         if (handle) {
 837                 if (close_session)
 838                         pam_code = pam_close_session(handle, PAM_DATA_SILENT);
 839
 840                 pam_end(handle, pam_code | PAM_DATA_SILENT);
 841         }
 842
 843         strv_free(e);
 844
 845         closelog();
 846
 847         if (pam_pid > 1) {
 848                 kill(pam_pid, SIGTERM);
 849                 kill(pam_pid, SIGCONT);
 850         }
 851
 852         return err;
 853 }
 854 #endif
 855
 856 static int do_capability_bounding_set_drop(uint64_t drop) {
 857         unsigned long i;
 858         cap_t old_cap = NULL, new_cap = NULL;
 859         cap_flag_value_t fv;
 860         int r;
 861
 862         /* If we are run as PID 1 we will lack CAP_SETPCAP by default
 863          * in the effective set (yes, the kernel drops that when
 864          * executing init!), so get it back temporarily so that we can
 865          * call PR_CAPBSET_DROP. */
 866
 867         old_cap = cap_get_proc();
 868         if (!old_cap)
 869                 return -errno;
 870
 871         if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
 872                 r = -errno;
 873                 goto finish;
 874         }
 875
 876         if (fv != CAP_SET) {
 877                 static const cap_value_t v = CAP_SETPCAP;
 878
 879                 new_cap = cap_dup(old_cap);
 880                 if (!new_cap) {
 881                         r = -errno;
 882                         goto finish;
 883                 }
 884
 885                 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
 886                         r = -errno;
 887                         goto finish;
 888                 }
 889
 890                 if (cap_set_proc(new_cap) < 0) {
 891                         r = -errno;
 892                         goto finish;
 893                 }
 894         }
 895
 896         for (i = 0; i <= cap_last_cap(); i++)
 897                 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
 898                         if (prctl(PR_CAPBSET_DROP, i) < 0) {
 899                                 r = -errno;
 900                                 goto finish;
 901                         }
 902                 }
 903
 904         r = 0;
 905
 906 finish:
 907         if (new_cap)
 908                 cap_free(new_cap);
 909
 910         if (old_cap) {
 911                 cap_set_proc(old_cap);
 912                 cap_free(old_cap);
 913         }
 914
 915         return r;
 916 }
 917
 918 int exec_spawn(ExecCommand *command,
 919                char **argv,
 920                const ExecContext *context,
 921                int fds[], unsigned n_fds,
 922                char **environment,
 923                bool apply_permissions,
 924                bool apply_chroot,
 925                bool apply_tty_stdin,
 926                bool confirm_spawn,
 927                CGroupBonding *cgroup_bondings,
 928                CGroupAttribute *cgroup_attributes,
 929                pid_t *ret) {
 930
 931         pid_t pid;
 932         int r;
 933         char *line;
 934         int socket_fd;
 935         char **files_env = NULL;
 936
 937         assert(command);
 938         assert(context);
 939         assert(ret);
 940         assert(fds || n_fds <= 0);
 941
 942         if (context->std_input == EXEC_INPUT_SOCKET ||
 943             context->std_output == EXEC_OUTPUT_SOCKET ||
 944             context->std_error == EXEC_OUTPUT_SOCKET) {
 945
 946                 if (n_fds != 1)
 947                         return -EINVAL;
 948
 949                 socket_fd = fds[0];
 950
 951                 fds = NULL;
 952                 n_fds = 0;
 953         } else
 954                 socket_fd = -1;
 955
 956         if ((r = exec_context_load_environment(context, &files_env)) < 0) {
 957                 log_error("Failed to load environment files: %s", strerror(-r));
 958                 return r;
 959         }
 960
 961         if (!argv)
 962                 argv = command->argv;
 963
 964         if (!(line = exec_command_line(argv))) {
 965                 r = -ENOMEM;
 966                 goto fail_parent;
 967         }
 968
 969         log_debug("About to execute: %s", line);
 970         free(line);
 971
 972         r = cgroup_bonding_realize_list(cgroup_bondings);
 973         if (r < 0)
 974                 goto fail_parent;
 975
 976         cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
 977
 978         if ((pid = fork()) < 0) {
 979                 r = -errno;
 980                 goto fail_parent;
 981         }
 982
 983         if (pid == 0) {
 984                 int i, err;
 985                 sigset_t ss;
 986                 const char *username = NULL, *home = NULL;
 987                 uid_t uid = (uid_t) -1;
 988                 gid_t gid = (gid_t) -1;
 989                 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
 990                 unsigned n_env = 0;
 991                 int saved_stdout = -1, saved_stdin = -1;
 992                 bool keep_stdout = false, keep_stdin = false;
 993
 994                 /* child */
 995
 996                 /* This string must fit in 10 chars (i.e. the length
 997                  * of "/sbin/init") */
 998                 rename_process("sd(EXEC)");
 999
1000                 /* We reset exactly these signals, since they are the
1001                  * only ones we set to SIG_IGN in the main daemon. All
1002                  * others we leave untouched because we set them to
1003                  * SIG_DFL or a valid handler initially, both of which
1004                  * will be demoted to SIG_DFL. */
1005                 default_signals(SIGNALS_CRASH_HANDLER,
1006                                 SIGNALS_IGNORE, -1);
1007
1008                 if (sigemptyset(&ss) < 0 ||
1009                     sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1010                         err = -errno;
1011                         r = EXIT_SIGNAL_MASK;
1012                         goto fail_child;
1013                 }
1014
1015                 /* Close sockets very early to make sure we don't
1016                  * block init reexecution because it cannot bind its
1017                  * sockets */
1018                 log_forget_fds();
1019                 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1020                                            socket_fd >= 0 ? 1 : n_fds);
1021                 if (err < 0) {
1022                         r = EXIT_FDS;
1023                         goto fail_child;
1024                 }
1025
1026                 if (!context->same_pgrp)
1027                         if (setsid() < 0) {
1028                                 err = -errno;
1029                                 r = EXIT_SETSID;
1030                                 goto fail_child;
1031                         }
1032
1033                 if (context->tcpwrap_name) {
1034                         if (socket_fd >= 0)
1035                                 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1036                                         err = -EACCES;
1037                                         r = EXIT_TCPWRAP;
1038                                         goto fail_child;
1039                                 }
1040
1041                         for (i = 0; i < (int) n_fds; i++) {
1042                                 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1043                                         err = -EACCES;
1044                                         r = EXIT_TCPWRAP;
1045                                         goto fail_child;
1046                                 }
1047                         }
1048                 }
1049
1050                 exec_context_tty_reset(context);
1051
1052                 /* We skip the confirmation step if we shall not apply the TTY */
1053                 if (confirm_spawn &&
1054                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1055                         char response;
1056
1057                         /* Set up terminal for the question */
1058                         if ((r = setup_confirm_stdio(context,
1059                                                      &saved_stdin, &saved_stdout))) {
1060                                 err = -errno;
1061                                 goto fail_child;
1062                         }
1063
1064                         /* Now ask the question. */
1065                         if (!(line = exec_command_line(argv))) {
1066                                 err = -ENOMEM;
1067                                 r = EXIT_MEMORY;
1068                                 goto fail_child;
1069                         }
1070
1071                         r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1072                         free(line);
1073
1074                         if (r < 0 || response == 'n') {
1075                                 err = -ECANCELED;
1076                                 r = EXIT_CONFIRM;
1077                                 goto fail_child;
1078                         } else if (response == 's') {
1079                                 err = r = 0;
1080                                 goto fail_child;
1081                         }
1082
1083                         /* Release terminal for the question */
1084                         if ((r = restore_confirm_stdio(context,
1085                                                        &saved_stdin, &saved_stdout,
1086                                                        &keep_stdin, &keep_stdout))) {
1087                                 err = -errno;
1088                                 goto fail_child;
1089                         }
1090                 }
1091
1092                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1093                  * must sure to drop O_NONBLOCK */
1094                 if (socket_fd >= 0)
1095                         fd_nonblock(socket_fd, false);
1096
1097                 if (!keep_stdin) {
1098                         err = setup_input(context, socket_fd, apply_tty_stdin);
1099                         if (err < 0) {
1100                                 r = EXIT_STDIN;
1101                                 goto fail_child;
1102                         }
1103                 }
1104
1105                 if (!keep_stdout) {
1106                         err = setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1107                         if (err < 0) {
1108                                 r = EXIT_STDOUT;
1109                                 goto fail_child;
1110                         }
1111                 }
1112
1113                 err = setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1114                 if (err < 0) {
1115                         r = EXIT_STDERR;
1116                         goto fail_child;
1117                 }
1118
1119                 if (cgroup_bondings) {
1120                         err = cgroup_bonding_install_list(cgroup_bondings, 0);
1121                         if (err < 0) {
1122                                 r = EXIT_CGROUP;
1123                                 goto fail_child;
1124                         }
1125                 }
1126
1127                 if (context->oom_score_adjust_set) {
1128                         char t[16];
1129
1130                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1131                         char_array_0(t);
1132
1133                         if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1134                                 /* Compatibility with Linux <= 2.6.35 */
1135
1136                                 int adj;
1137
1138                                 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1139                                 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1140
1141                                 snprintf(t, sizeof(t), "%i", adj);
1142                                 char_array_0(t);
1143
1144                                 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1145                                     && errno != EACCES) {
1146                                         err = -errno;
1147                                         r = EXIT_OOM_ADJUST;
1148                                         goto fail_child;
1149                                 }
1150                         }
1151                 }
1152
1153                 if (context->nice_set)
1154                         if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1155                                 err = -errno;
1156                                 r = EXIT_NICE;
1157                                 goto fail_child;
1158                         }
1159
1160                 if (context->cpu_sched_set) {
1161                         struct sched_param param;
1162
1163                         zero(param);
1164                         param.sched_priority = context->cpu_sched_priority;
1165
1166                         if (sched_setscheduler(0, context->cpu_sched_policy |
1167                                                (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
1168                                 err = -errno;
1169                                 r = EXIT_SETSCHEDULER;
1170                                 goto fail_child;
1171                         }
1172                 }
1173
1174                 if (context->cpuset)
1175                         if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1176                                 err = -errno;
1177                                 r = EXIT_CPUAFFINITY;
1178                                 goto fail_child;
1179                         }
1180
1181                 if (context->ioprio_set)
1182                         if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1183                                 err = -errno;
1184                                 r = EXIT_IOPRIO;
1185                                 goto fail_child;
1186                         }
1187
1188                 if (context->timer_slack_nsec_set)
1189                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1190                                 err = -errno;
1191                                 r = EXIT_TIMERSLACK;
1192                                 goto fail_child;
1193                         }
1194
1195                 if (context->utmp_id)
1196                         utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1197
1198                 if (context->user) {
1199                         username = context->user;
1200                         err = get_user_creds(&username, &uid, &gid, &home);
1201                         if (err < 0) {
1202                                 r = EXIT_USER;
1203                                 goto fail_child;
1204                         }
1205
1206                         if (is_terminal_input(context->std_input)) {
1207                                 err = chown_terminal(STDIN_FILENO, uid);
1208                                 if (err < 0) {
1209                                         r = EXIT_STDIN;
1210                                         goto fail_child;
1211                                 }
1212                         }
1213
1214                         if (cgroup_bondings && context->control_group_modify) {
1215                                 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1216                                 if (err >= 0)
1217                                         err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid);
1218                                 if (err < 0) {
1219                                         r = EXIT_CGROUP;
1220                                         goto fail_child;
1221                                 }
1222                         }
1223                 }
1224
1225                 if (apply_permissions) {
1226                         err = enforce_groups(context, username, gid);
1227                         if (err < 0) {
1228                                 r = EXIT_GROUP;
1229                                 goto fail_child;
1230                         }
1231                 }
1232
1233                 umask(context->umask);
1234
1235 #ifdef HAVE_PAM
1236                 if (context->pam_name && username) {
1237                         err = setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds);
1238                         if (err < 0) {
1239                                 r = EXIT_PAM;
1240                                 goto fail_child;
1241                         }
1242                 }
1243 #endif
1244                 if (context->private_network) {
1245                         if (unshare(CLONE_NEWNET) < 0) {
1246                                 err = -errno;
1247                                 r = EXIT_NETWORK;
1248                                 goto fail_child;
1249                         }
1250
1251                         loopback_setup();
1252                 }
1253
1254                 if (strv_length(context->read_write_dirs) > 0 ||
1255                     strv_length(context->read_only_dirs) > 0 ||
1256                     strv_length(context->inaccessible_dirs) > 0 ||
1257                     context->mount_flags != MS_SHARED ||
1258                     context->private_tmp) {
1259                         err = setup_namespace(context->read_write_dirs,
1260                                               context->read_only_dirs,
1261                                               context->inaccessible_dirs,
1262                                               context->private_tmp,
1263                                               context->mount_flags);
1264                         if (err < 0) {
1265                                 r = EXIT_NAMESPACE;
1266                                 goto fail_child;
1267                         }
1268                 }
1269
1270                 if (apply_chroot) {
1271                         if (context->root_directory)
1272                                 if (chroot(context->root_directory) < 0) {
1273                                         err = -errno;
1274                                         r = EXIT_CHROOT;
1275                                         goto fail_child;
1276                                 }
1277
1278                         if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1279                                 err = -errno;
1280                                 r = EXIT_CHDIR;
1281                                 goto fail_child;
1282                         }
1283                 } else {
1284
1285                         char *d;
1286
1287                         if (asprintf(&d, "%s/%s",
1288                                      context->root_directory ? context->root_directory : "",
1289                                      context->working_directory ? context->working_directory : "") < 0) {
1290                                 err = -ENOMEM;
1291                                 r = EXIT_MEMORY;
1292                                 goto fail_child;
1293                         }
1294
1295                         if (chdir(d) < 0) {
1296                                 err = -errno;
1297                                 free(d);
1298                                 r = EXIT_CHDIR;
1299                                 goto fail_child;
1300                         }
1301
1302                         free(d);
1303                 }
1304
1305                 /* We repeat the fd closing here, to make sure that
1306                  * nothing is leaked from the PAM modules */
1307                 err = close_all_fds(fds, n_fds);
1308                 if (err >= 0)
1309                         err = shift_fds(fds, n_fds);
1310                 if (err >= 0)
1311                         err = flags_fds(fds, n_fds, context->non_blocking);
1312                 if (err < 0) {
1313                         r = EXIT_FDS;
1314                         goto fail_child;
1315                 }
1316
1317                 if (apply_permissions) {
1318
1319                         for (i = 0; i < RLIMIT_NLIMITS; i++) {
1320                                 if (!context->rlimit[i])
1321                                         continue;
1322
1323                                 if (setrlimit(i, context->rlimit[i]) < 0) {
1324                                         err = -errno;
1325                                         r = EXIT_LIMITS;
1326                                         goto fail_child;
1327                                 }
1328                         }
1329
1330                         if (context->capability_bounding_set_drop) {
1331                                 err = do_capability_bounding_set_drop(context->capability_bounding_set_drop);
1332                                 if (err < 0) {
1333                                         r = EXIT_CAPABILITIES;
1334                                         goto fail_child;
1335                                 }
1336                         }
1337
1338                         if (context->user) {
1339                                 err = enforce_user(context, uid);
1340                                 if (err < 0) {
1341                                         r = EXIT_USER;
1342                                         goto fail_child;
1343                                 }
1344                         }
1345
1346                         /* PR_GET_SECUREBITS is not privileged, while
1347                          * PR_SET_SECUREBITS is. So to suppress
1348                          * potential EPERMs we'll try not to call
1349                          * PR_SET_SECUREBITS unless necessary. */
1350                         if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1351                                 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1352                                         err = -errno;
1353                                         r = EXIT_SECUREBITS;
1354                                         goto fail_child;
1355                                 }
1356
1357                         if (context->capabilities)
1358                                 if (cap_set_proc(context->capabilities) < 0) {
1359                                         err = -errno;
1360                                         r = EXIT_CAPABILITIES;
1361                                         goto fail_child;
1362                                 }
1363                 }
1364
1365                 if (!(our_env = new0(char*, 7))) {
1366                         err = -ENOMEM;
1367                         r = EXIT_MEMORY;
1368                         goto fail_child;
1369                 }
1370
1371                 if (n_fds > 0)
1372                         if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1373                             asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1374                                 err = -ENOMEM;
1375                                 r = EXIT_MEMORY;
1376                                 goto fail_child;
1377                         }
1378
1379                 if (home)
1380                         if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1381                                 err = -ENOMEM;
1382                                 r = EXIT_MEMORY;
1383                                 goto fail_child;
1384                         }
1385
1386                 if (username)
1387                         if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1388                             asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1389                                 err = -ENOMEM;
1390                                 r = EXIT_MEMORY;
1391                                 goto fail_child;
1392                         }
1393
1394                 if (is_terminal_input(context->std_input) ||
1395                     context->std_output == EXEC_OUTPUT_TTY ||
1396                     context->std_error == EXEC_OUTPUT_TTY)
1397                         if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1398                                 err = -ENOMEM;
1399                                 r = EXIT_MEMORY;
1400                                 goto fail_child;
1401                         }
1402
1403                 assert(n_env <= 7);
1404
1405                 if (!(final_env = strv_env_merge(
1406                                       5,
1407                                       environment,
1408                                       our_env,
1409                                       context->environment,
1410                                       files_env,
1411                                       pam_env,
1412                                       NULL))) {
1413                         err = -ENOMEM;
1414                         r = EXIT_MEMORY;
1415                         goto fail_child;
1416                 }
1417
1418                 if (!(final_argv = replace_env_argv(argv, final_env))) {
1419                         err = -ENOMEM;
1420                         r = EXIT_MEMORY;
1421                         goto fail_child;
1422                 }
1423
1424                 final_env = strv_env_clean(final_env);
1425
1426                 execve(command->path, final_argv, final_env);
1427                 err = -errno;
1428                 r = EXIT_EXEC;
1429
1430         fail_child:
1431                 if (r != 0) {
1432                         log_open();
1433                         log_warning("Failed at step %s spawning %s: %s",
1434                                     exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1435                                     command->path, strerror(-err));
1436                 }
1437
1438                 strv_free(our_env);
1439                 strv_free(final_env);
1440                 strv_free(pam_env);
1441                 strv_free(files_env);
1442                 strv_free(final_argv);
1443
1444                 if (saved_stdin >= 0)
1445                         close_nointr_nofail(saved_stdin);
1446
1447                 if (saved_stdout >= 0)
1448                         close_nointr_nofail(saved_stdout);
1449
1450                 _exit(r);
1451         }
1452
1453         strv_free(files_env);
1454
1455         /* We add the new process to the cgroup both in the child (so
1456          * that we can be sure that no user code is ever executed
1457          * outside of the cgroup) and in the parent (so that we can be
1458          * sure that when we kill the cgroup the process will be
1459          * killed too). */
1460         if (cgroup_bondings)
1461                 cgroup_bonding_install_list(cgroup_bondings, pid);
1462
1463         log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1464
1465         exec_status_start(&command->exec_status, pid);
1466
1467         *ret = pid;
1468         return 0;
1469
1470 fail_parent:
1471         strv_free(files_env);
1472
1473         return r;
1474 }
1475
1476 void exec_context_init(ExecContext *c) {
1477         assert(c);
1478
1479         c->umask = 0022;
1480         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1481         c->cpu_sched_policy = SCHED_OTHER;
1482         c->syslog_priority = LOG_DAEMON|LOG_INFO;
1483         c->syslog_level_prefix = true;
1484         c->mount_flags = MS_SHARED;
1485         c->kill_signal = SIGTERM;
1486         c->send_sigkill = true;
1487 }
1488
1489 void exec_context_done(ExecContext *c) {
1490         unsigned l;
1491
1492         assert(c);
1493
1494         strv_free(c->environment);
1495         c->environment = NULL;
1496
1497         strv_free(c->environment_files);
1498         c->environment_files = NULL;
1499
1500         for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1501                 free(c->rlimit[l]);
1502                 c->rlimit[l] = NULL;
1503         }
1504
1505         free(c->working_directory);
1506         c->working_directory = NULL;
1507         free(c->root_directory);
1508         c->root_directory = NULL;
1509
1510         free(c->tty_path);
1511         c->tty_path = NULL;
1512
1513         free(c->tcpwrap_name);
1514         c->tcpwrap_name = NULL;
1515
1516         free(c->syslog_identifier);
1517         c->syslog_identifier = NULL;
1518
1519         free(c->user);
1520         c->user = NULL;
1521
1522         free(c->group);
1523         c->group = NULL;
1524
1525         strv_free(c->supplementary_groups);
1526         c->supplementary_groups = NULL;
1527
1528         free(c->pam_name);
1529         c->pam_name = NULL;
1530
1531         if (c->capabilities) {
1532                 cap_free(c->capabilities);
1533                 c->capabilities = NULL;
1534         }
1535
1536         strv_free(c->read_only_dirs);
1537         c->read_only_dirs = NULL;
1538
1539         strv_free(c->read_write_dirs);
1540         c->read_write_dirs = NULL;
1541
1542         strv_free(c->inaccessible_dirs);
1543         c->inaccessible_dirs = NULL;
1544
1545         if (c->cpuset)
1546                 CPU_FREE(c->cpuset);
1547
1548         free(c->utmp_id);
1549         c->utmp_id = NULL;
1550 }
1551
1552 void exec_command_done(ExecCommand *c) {
1553         assert(c);
1554
1555         free(c->path);
1556         c->path = NULL;
1557
1558         strv_free(c->argv);
1559         c->argv = NULL;
1560 }
1561
1562 void exec_command_done_array(ExecCommand *c, unsigned n) {
1563         unsigned i;
1564
1565         for (i = 0; i < n; i++)
1566                 exec_command_done(c+i);
1567 }
1568
1569 void exec_command_free_list(ExecCommand *c) {
1570         ExecCommand *i;
1571
1572         while ((i = c)) {
1573                 LIST_REMOVE(ExecCommand, command, c, i);
1574                 exec_command_done(i);
1575                 free(i);
1576         }
1577 }
1578
1579 void exec_command_free_array(ExecCommand **c, unsigned n) {
1580         unsigned i;
1581
1582         for (i = 0; i < n; i++) {
1583                 exec_command_free_list(c[i]);
1584                 c[i] = NULL;
1585         }
1586 }
1587
1588 int exec_context_load_environment(const ExecContext *c, char ***l) {
1589         char **i, **r = NULL;
1590
1591         assert(c);
1592         assert(l);
1593
1594         STRV_FOREACH(i, c->environment_files) {
1595                 char *fn;
1596                 int k;
1597                 bool ignore = false;
1598                 char **p;
1599
1600                 fn = *i;
1601
1602                 if (fn[0] == '-') {
1603                         ignore = true;
1604                         fn ++;
1605                 }
1606
1607                 if (!path_is_absolute(fn)) {
1608
1609                         if (ignore)
1610                                 continue;
1611
1612                         strv_free(r);
1613                         return -EINVAL;
1614                 }
1615
1616                 if ((k = load_env_file(fn, &p)) < 0) {
1617
1618                         if (ignore)
1619                                 continue;
1620
1621                         strv_free(r);
1622                         return k;
1623                 }
1624
1625                 if (r == NULL)
1626                         r = p;
1627                 else {
1628                         char **m;
1629
1630                         m = strv_env_merge(2, r, p);
1631                         strv_free(r);
1632                         strv_free(p);
1633
1634                         if (!m)
1635                                 return -ENOMEM;
1636
1637                         r = m;
1638                 }
1639         }
1640
1641         *l = r;
1642
1643         return 0;
1644 }
1645
1646 static void strv_fprintf(FILE *f, char **l) {
1647         char **g;
1648
1649         assert(f);
1650
1651         STRV_FOREACH(g, l)
1652                 fprintf(f, " %s", *g);
1653 }
1654
1655 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1656         char ** e;
1657         unsigned i;
1658
1659         assert(c);
1660         assert(f);
1661
1662         if (!prefix)
1663                 prefix = "";
1664
1665         fprintf(f,
1666                 "%sUMask: %04o\n"
1667                 "%sWorkingDirectory: %s\n"
1668                 "%sRootDirectory: %s\n"
1669                 "%sNonBlocking: %s\n"
1670                 "%sPrivateTmp: %s\n"
1671                 "%sControlGroupModify: %s\n"
1672                 "%sPrivateNetwork: %s\n",
1673                 prefix, c->umask,
1674                 prefix, c->working_directory ? c->working_directory : "/",
1675                 prefix, c->root_directory ? c->root_directory : "/",
1676                 prefix, yes_no(c->non_blocking),
1677                 prefix, yes_no(c->private_tmp),
1678                 prefix, yes_no(c->control_group_modify),
1679                 prefix, yes_no(c->private_network));
1680
1681         STRV_FOREACH(e, c->environment)
1682                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1683
1684         STRV_FOREACH(e, c->environment_files)
1685                 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1686
1687         if (c->tcpwrap_name)
1688                 fprintf(f,
1689                         "%sTCPWrapName: %s\n",
1690                         prefix, c->tcpwrap_name);
1691
1692         if (c->nice_set)
1693                 fprintf(f,
1694                         "%sNice: %i\n",
1695                         prefix, c->nice);
1696
1697         if (c->oom_score_adjust_set)
1698                 fprintf(f,
1699                         "%sOOMScoreAdjust: %i\n",
1700                         prefix, c->oom_score_adjust);
1701
1702         for (i = 0; i < RLIM_NLIMITS; i++)
1703                 if (c->rlimit[i])
1704                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1705
1706         if (c->ioprio_set)
1707                 fprintf(f,
1708                         "%sIOSchedulingClass: %s\n"
1709                         "%sIOPriority: %i\n",
1710                         prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1711                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1712
1713         if (c->cpu_sched_set)
1714                 fprintf(f,
1715                         "%sCPUSchedulingPolicy: %s\n"
1716                         "%sCPUSchedulingPriority: %i\n"
1717                         "%sCPUSchedulingResetOnFork: %s\n",
1718                         prefix, sched_policy_to_string(c->cpu_sched_policy),
1719                         prefix, c->cpu_sched_priority,
1720                         prefix, yes_no(c->cpu_sched_reset_on_fork));
1721
1722         if (c->cpuset) {
1723                 fprintf(f, "%sCPUAffinity:", prefix);
1724                 for (i = 0; i < c->cpuset_ncpus; i++)
1725                         if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1726                                 fprintf(f, " %i", i);
1727                 fputs("\n", f);
1728         }
1729
1730         if (c->timer_slack_nsec_set)
1731                 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1732
1733         fprintf(f,
1734                 "%sStandardInput: %s\n"
1735                 "%sStandardOutput: %s\n"
1736                 "%sStandardError: %s\n",
1737                 prefix, exec_input_to_string(c->std_input),
1738                 prefix, exec_output_to_string(c->std_output),
1739                 prefix, exec_output_to_string(c->std_error));
1740
1741         if (c->tty_path)
1742                 fprintf(f,
1743                         "%sTTYPath: %s\n"
1744                         "%sTTYReset: %s\n"
1745                         "%sTTYVHangup: %s\n"
1746                         "%sTTYVTDisallocate: %s\n",
1747                         prefix, c->tty_path,
1748                         prefix, yes_no(c->tty_reset),
1749                         prefix, yes_no(c->tty_vhangup),
1750                         prefix, yes_no(c->tty_vt_disallocate));
1751
1752         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1753             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1754             c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1755             c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1756                 fprintf(f,
1757                         "%sSyslogFacility: %s\n"
1758                         "%sSyslogLevel: %s\n",
1759                         prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1760                         prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1761
1762         if (c->capabilities) {
1763                 char *t;
1764                 if ((t = cap_to_text(c->capabilities, NULL))) {
1765                         fprintf(f, "%sCapabilities: %s\n",
1766                                 prefix, t);
1767                         cap_free(t);
1768                 }
1769         }
1770
1771         if (c->secure_bits)
1772                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1773                         prefix,
1774                         (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1775                         (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1776                         (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1777                         (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1778                         (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1779                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1780
1781         if (c->capability_bounding_set_drop) {
1782                 unsigned long l;
1783                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1784
1785                 for (l = 0; l <= cap_last_cap(); l++)
1786                         if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1787                                 char *t;
1788
1789                                 if ((t = cap_to_name(l))) {
1790                                         fprintf(f, " %s", t);
1791                                         cap_free(t);
1792                                 }
1793                         }
1794
1795                 fputs("\n", f);
1796         }
1797
1798         if (c->user)
1799                 fprintf(f, "%sUser: %s\n", prefix, c->user);
1800         if (c->group)
1801                 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1802
1803         if (strv_length(c->supplementary_groups) > 0) {
1804                 fprintf(f, "%sSupplementaryGroups:", prefix);
1805                 strv_fprintf(f, c->supplementary_groups);
1806                 fputs("\n", f);
1807         }
1808
1809         if (c->pam_name)
1810                 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1811
1812         if (strv_length(c->read_write_dirs) > 0) {
1813                 fprintf(f, "%sReadWriteDirs:", prefix);
1814                 strv_fprintf(f, c->read_write_dirs);
1815                 fputs("\n", f);
1816         }
1817
1818         if (strv_length(c->read_only_dirs) > 0) {
1819                 fprintf(f, "%sReadOnlyDirs:", prefix);
1820                 strv_fprintf(f, c->read_only_dirs);
1821                 fputs("\n", f);
1822         }
1823
1824         if (strv_length(c->inaccessible_dirs) > 0) {
1825                 fprintf(f, "%sInaccessibleDirs:", prefix);
1826                 strv_fprintf(f, c->inaccessible_dirs);
1827                 fputs("\n", f);
1828         }
1829
1830         fprintf(f,
1831                 "%sKillMode: %s\n"
1832                 "%sKillSignal: SIG%s\n"
1833                 "%sSendSIGKILL: %s\n",
1834                 prefix, kill_mode_to_string(c->kill_mode),
1835                 prefix, signal_to_string(c->kill_signal),
1836                 prefix, yes_no(c->send_sigkill));
1837
1838         if (c->utmp_id)
1839                 fprintf(f,
1840                         "%sUtmpIdentifier: %s\n",
1841                         prefix, c->utmp_id);
1842 }
1843
1844 void exec_status_start(ExecStatus *s, pid_t pid) {
1845         assert(s);
1846
1847         zero(*s);
1848         s->pid = pid;
1849         dual_timestamp_get(&s->start_timestamp);
1850 }
1851
1852 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1853         assert(s);
1854
1855         if (s->pid && s->pid != pid)
1856                 zero(*s);
1857
1858         s->pid = pid;
1859         dual_timestamp_get(&s->exit_timestamp);
1860
1861         s->code = code;
1862         s->status = status;
1863
1864         if (context) {
1865                 if (context->utmp_id)
1866                         utmp_put_dead_process(context->utmp_id, pid, code, status);
1867
1868                 exec_context_tty_reset(context);
1869         }
1870 }
1871
1872 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1873         char buf[FORMAT_TIMESTAMP_MAX];
1874
1875         assert(s);
1876         assert(f);
1877
1878         if (!prefix)
1879                 prefix = "";
1880
1881         if (s->pid <= 0)
1882                 return;
1883
1884         fprintf(f,
1885                 "%sPID: %lu\n",
1886                 prefix, (unsigned long) s->pid);
1887
1888         if (s->start_timestamp.realtime > 0)
1889                 fprintf(f,
1890                         "%sStart Timestamp: %s\n",
1891                         prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1892
1893         if (s->exit_timestamp.realtime > 0)
1894                 fprintf(f,
1895                         "%sExit Timestamp: %s\n"
1896                         "%sExit Code: %s\n"
1897                         "%sExit Status: %i\n",
1898                         prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1899                         prefix, sigchld_code_to_string(s->code),
1900                         prefix, s->status);
1901 }
1902
1903 char *exec_command_line(char **argv) {
1904         size_t k;
1905         char *n, *p, **a;
1906         bool first = true;
1907
1908         assert(argv);
1909
1910         k = 1;
1911         STRV_FOREACH(a, argv)
1912                 k += strlen(*a)+3;
1913
1914         if (!(n = new(char, k)))
1915                 return NULL;
1916
1917         p = n;
1918         STRV_FOREACH(a, argv) {
1919
1920                 if (!first)
1921                         *(p++) = ' ';
1922                 else
1923                         first = false;
1924
1925                 if (strpbrk(*a, WHITESPACE)) {
1926                         *(p++) = '\'';
1927                         p = stpcpy(p, *a);
1928                         *(p++) = '\'';
1929                 } else
1930                         p = stpcpy(p, *a);
1931
1932         }
1933
1934         *p = 0;
1935
1936         /* FIXME: this doesn't really handle arguments that have
1937          * spaces and ticks in them */
1938
1939         return n;
1940 }
1941
1942 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1943         char *p2;
1944         const char *prefix2;
1945
1946         char *cmd;
1947
1948         assert(c);
1949         assert(f);
1950
1951         if (!prefix)
1952                 prefix = "";
1953         p2 = strappend(prefix, "\t");
1954         prefix2 = p2 ? p2 : prefix;
1955
1956         cmd = exec_command_line(c->argv);
1957
1958         fprintf(f,
1959                 "%sCommand Line: %s\n",
1960                 prefix, cmd ? cmd : strerror(ENOMEM));
1961
1962         free(cmd);
1963
1964         exec_status_dump(&c->exec_status, f, prefix2);
1965
1966         free(p2);
1967 }
1968
1969 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1970         assert(f);
1971
1972         if (!prefix)
1973                 prefix = "";
1974
1975         LIST_FOREACH(command, c, c)
1976                 exec_command_dump(c, f, prefix);
1977 }
1978
1979 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1980         ExecCommand *end;
1981
1982         assert(l);
1983         assert(e);
1984
1985         if (*l) {
1986                 /* It's kind of important, that we keep the order here */
1987                 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1988                 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1989         } else
1990               *l = e;
1991 }
1992
1993 int exec_command_set(ExecCommand *c, const char *path, ...) {
1994         va_list ap;
1995         char **l, *p;
1996
1997         assert(c);
1998         assert(path);
1999
2000         va_start(ap, path);
2001         l = strv_new_ap(path, ap);
2002         va_end(ap);
2003
2004         if (!l)
2005                 return -ENOMEM;
2006
2007         if (!(p = strdup(path))) {
2008                 strv_free(l);
2009                 return -ENOMEM;
2010         }
2011
2012         free(c->path);
2013         c->path = p;
2014
2015         strv_free(c->argv);
2016         c->argv = l;
2017
2018         return 0;
2019 }
2020
2021 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2022         [EXEC_INPUT_NULL] = "null",
2023         [EXEC_INPUT_TTY] = "tty",
2024         [EXEC_INPUT_TTY_FORCE] = "tty-force",
2025         [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2026         [EXEC_INPUT_SOCKET] = "socket"
2027 };
2028
2029 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2030
2031 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2032         [EXEC_OUTPUT_INHERIT] = "inherit",
2033         [EXEC_OUTPUT_NULL] = "null",
2034         [EXEC_OUTPUT_TTY] = "tty",
2035         [EXEC_OUTPUT_SYSLOG] = "syslog",
2036         [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2037         [EXEC_OUTPUT_KMSG] = "kmsg",
2038         [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2039         [EXEC_OUTPUT_SOCKET] = "socket"
2040 };
2041
2042 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2043
2044 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2045         [KILL_CONTROL_GROUP] = "control-group",
2046         [KILL_PROCESS] = "process",
2047         [KILL_NONE] = "none"
2048 };
2049
2050 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2051
2052 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2053         [KILL_MAIN] = "main",
2054         [KILL_CONTROL] = "control",
2055         [KILL_ALL] = "all"
2056 };
2057
2058 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);